A Hybrid Machine Learning Method for Fusing fMRI and Genetic Data: Combining both Improves Classification of Schizophrenia

Honghui Yang, Jingyu Liu, Jing Sui, Godfrey D. Pearlson, Vince D. CalhounFrontiers in Human Neuroscience 2010 · DOI

Abstract

We demonstrate a hybrid machine learning method to classify schizophrenia patients and healthy controls, using functional magnetic resonance imaging (fMRI) and single nucleotide polymorphism (SNP) data. The method consists of four stages: (1) SNPs with the most discriminating information between the healthy controls and schizophrenia patients are selected to construct a support vector machine ensemble (SNP-SVME). (2) Voxels in the fMRI map contributing to classification are selected to build another SVME (Voxel-SVME). (3) Components of fMRI activation obtained with independent component analysis (ICA) are used to construct a single SVM classifier (ICA-SVMC). (4) The above three models are combined into a single module using a majority voting approach to make a final decision (Combined SNP-fMRI). The method was evaluated by a fully validated leave-one-out method using 40 subjects (20 patients and 20 controls). The classification accuracy was: 0.74 for SNP-SVME, 0.82 for Voxel-SVME, 0.83 for ICA-SVMC, and 0.87 for Combined SNP-fMRI. Experimental results show that better classification accuracy was achieved by combining genetic and fMRI data than using either alone, indicating that genetic and brain function representing different, but partially complementary aspects, of schizophrenia etiopathology.

Introduction

Schizophrenia is a complex, chronic brain disorder whose diagnosis currently relies heavily on clinical examinations and subjective observations of symptoms. While biological markers exist in both genetics and brain imaging, these fields often operate in silos. Integrating these data sources presents a significant technical challenge, primarily due to the "Small N, Large P" problem: medical studies often involve a small number of subjects ($N$) but thousands of features ($P$), such as voxels or genetic markers, making models highly prone to overfitting.

Methodology

To address these challenges, we developed a four-stage hybrid ensemble approach designed to fuse categorical genetic data with high-dimensional, spatial-temporal fMRI data.

Flow chart of the proposed hybrid machine learning method for data fusion.
  1. SNP-SVME (Genetic Branch): Single Nucleotide Polymorphisms (SNPs) were filtered using Forward Sequential Feature Selection (FSFS) to remove noise. A Feature Selective AdaBoost ensemble of Support Vector Machines (SVMs) was then constructed to classify subjects based on their genetic profile.
  2. Voxel-SVME (Spatial fMRI Branch): To manage the dimensionality of fMRI scans, neighboring voxels were averaged to reduce the search space before applying an AdaBoost-SVM ensemble to identify predictive brain regions.
  3. ICA-SVMC (Connectivity fMRI Branch): Independent Component Analysis (ICA) was used to identify hidden patterns of brain activation, which were then classified using a standard linear SVM.
  4. Majority Voting: The final diagnosis was determined by a majority vote across the three previous stages, ensuring a robust final decision based on multiple biological "views."

Experimental Results

The study utilized a dataset of 40 subjects (20 patients and 20 controls). Genetic data included 367 SNPs from 222 genes related to neurobiology, while fMRI data were captured during an auditory oddball task.

The hybrid approach demonstrated a clear synergistic effect. Individual model accuracies were 74% for SNPs, 82% for voxels, and 83% for ICA components. However, the Combined (Hybrid) model achieved 87% accuracy, with a sensitivity of 85.75% and specificity of 88.75%.

The importance of individual SNPs in the classification process, highlighting the effectiveness of the feature selection stage.

Crucially, the feature selection process was essential; without it, accuracy dropped significantly (as low as 40% for SNPs), proving that the model successfully filtered out biological noise to focus on relevant markers.

Discussion and Conclusion

The primary insight of this research is the complementarity of fMRI and genetic data. While fMRI is "closer" to the actual expression of symptoms and generally yielded higher individual accuracy, the genetic model (SNP-SVME) often provided the critical tie-breaking insight needed for a correct diagnosis when imaging data was ambiguous.

Furthermore, the model identified biologically relevant markers, including well-known schizophrenia-related genes (such as COMT and DISC1) and specific brain regions (such as the frontal gyri and cingulate gyrus). While limited by a small sample size, this study suggests an effective way to reassess biological classification for schizophrenia, potentially identifying diagnostically important markers for the disorder.

Summary updated

Abstract

We demonstrate a hybrid machine learning method to classify schizophrenia patients and healthy controls, using functional magnetic resonance imaging (fMRI) and single nucleotide polymorphism (SNP) data. The method consists of four stages: (1) SNPs with the most discriminating information between the healthy controls and schizophrenia patients are selected to construct a support vector machine ensemble (SNP-SVME). (2) Voxels in the fMRI map contributing to classification are selected to build another SVME (Voxel-SVME). (3) Components of fMRI activation obtained with independent component analysis (ICA) are used to construct a single SVM classifier (ICA-SVMC). (4) The above three models are combined into a single module using a majority voting approach to make a final decision (Combined SNP-fMRI). The method was evaluated by a fully validated leave-one-out method using 40 subjects (20 patients and 20 controls). The classification accuracy was: 0.74 for SNP-SVME, 0.82 for Voxel-SVME, 0.83 for ICA-SVMC, and 0.87 for Combined SNP-fMRI. Experimental results show that better classification accuracy was achieved by combining genetic and fMRI data than using either alone, indicating that genetic and brain function representing different, but partially complementary aspects, of schizophrenia etiopathology. This study suggests an effective way to reassess biological classification of individuals with schizophrenia, which is also potentially useful for identifying diagnostically important markers for the disorder.
BibTeX
@inproceedings{yang2010hybrid,
  author = {Honghui Yang and Jingyu Liu and Jing Sui and Godfrey D. Pearlson and Vince D. Calhoun},
  title = {{A Hybrid Machine Learning Method for Fusing fMRI and Genetic Data: Combining both Improves Classification of Schizophrenia}},
  booktitle = {Frontiers in Human Neuroscience},
  year = {2010},
  doi = {10.3389/fnhum.2010.00192},
}
Figure